Describe the workflow you want to enable

Naive Bayes classifiers (GaussianNB, MultinomialNB, BernoulliNB, ComplementNB, CategoricalNB) currently do not support the class_weight parameter, while almost all other scikit-learn classifiers do. This creates API inconsistency and requires users to manually compute and pass sample_weight for imbalanced datasets.

Current Workaround

from sklearn.naive_bayes import GaussianNB
from sklearn.utils.class_weight import compute_sample_weight
sample_weight = compute_sample_weight('balanced', y)
gnb = GaussianNB()
gnb.fit(X, y, sample_weight=sample_weight)

Proposed Enhancement

gnb = GaussianNB(class_weight='balanced')
gnb.fit(X, y)

Describe your proposed solution

Add a class_weight parameter to all Naive Bayes classifiers that internally converts class weights to sample weights before calling the existing fit() logic. This leverages the fact that all NB classifiers already support sample_weight in their fitting methods.

class GaussianNB(_BaseNB):
    _parameter_constraints: dict = {
        "priors": ["array-like", None],
        "var_smoothing": [Interval(Real, 0, None, closed="left")],
        "class_weight": ["dict", "balanced", None],  # NEW
    }

    def __init__(self, *, priors=None, var_smoothing=1e-9, class_weight=None):
        self.priors = priors
        self.var_smoothing = var_smoothing
-->  self.class_weight = class_weight 

Describe alternatives you've considered, if relevant

No response

Additional context

This is a improvement for API consistency. While the workaround exists, having uniform behavior across classifiers reduces confusion and improves discoverability, especially for beginners.